OpenAI to Build $30 Billion 'Project Camellia' Data Center in Georgia — Its First Self-Designed Campus

OpenAI has unveiled its most ambitious infrastructure project yet: a data center campus in Effingham County, Georgia, that will consume 3.2 gigawatts of electricity at full capacity and cost well north of $30 billion to build out — enough power to theoretically supply roughly 2.4 million U.S. homes, and an investment that rivals the GDP of some small nations.

Dubbed Project Camellia, the facility represents a qualitative shift in OpenAI’s infrastructure posture. Until now, the company has relied on partnerships with Microsoft Azure and other hyperscale cloud providers to supply the computational horsepower it needs to train and serve its models. Project Camellia is its first attempt to design and build compute infrastructure from scratch — owning the land, the buildings, the power contracts, and the cooling systems rather than renting them.

Scale and Location

The site sits in Effingham County, near Savannah, Georgia, at a location known as the Savannah Gateway Industrial Hub. The 1,400-acre campus places OpenAI roughly 45 minutes from one of the U.S. Southeast’s largest port cities, with access to Georgia Power’s grid under a contract securing the full 3.2 GW allocation.

For context, 3.2 gigawatts is more than twice the peak capacity of the entire Hoover Dam. It exceeds the total computing power of any single data center complex currently in public operation. At current GPU power densities, a campus of this scale could potentially host hundreds of thousands of next-generation AI accelerators operating simultaneously.

The project will be built in phases. Sachin Katti, OpenAI’s vice president of compute strategy, said the company expects “several hundred megawatts” of capacity to come online starting in 2028, with the full buildout continuing through 2032. OpenAI is also actively seeking infrastructure and financial partners to share the construction burden.

The $750 Billion Roadmap

Project Camellia does not exist in isolation. It is the most visible hardware expression of a dramatically expanded compute ambition. OpenAI has revised its projected infrastructure spending through 2030 upward to $750 billion, compared to the roughly $600 billion figure cited earlier in 2026.

The revision reflects a simple internal calculus: the more capable OpenAI’s models become, the more inference demand they generate, and the more training compute the next generation requires. In a model-scaling regime where capability advances track closely with compute investment, the only way to maintain the frontier is to keep spending.

The Georgia campus will serve both training and inference workloads, but the emphasis in near-term phases is likely to be inference — serving the growing number of users and enterprise customers accessing OpenAI’s models via API and through applications like ChatGPT.

Why Self-Build?

OpenAI’s decision to become its own data center developer, rather than continuing to rely on hyperscale partners, reflects several pressures converging simultaneously.

First, there is the question of power access. The most capable AI clusters require custom power delivery, cooling, and network topology that is difficult to achieve in shared cloud environments. Building from the ground up allows OpenAI to optimize its physical infrastructure for AI workloads in ways that off-the-shelf cloud data centers cannot.

Second, there is cost. Renting compute from a hyperscaler like Microsoft Azure carries significant markup over owning equivalent hardware. At the scale OpenAI now operates — and the scale it is projecting — the economics of ownership eventually overwhelm the benefits of outsourcing.

Third, there is strategic independence. Microsoft has historically been OpenAI’s primary compute partner under a long-term agreement, but that relationship has become increasingly complex as Microsoft develops and promotes its own competing MAI family of models. Owning proprietary infrastructure reduces OpenAI’s dependence on a partner that is simultaneously its largest investor and a direct competitor in the enterprise AI market.

Navigating Local Opposition

The announcement was not without friction. Utility rate increases and environmental concerns associated with the AI data center buildout wave have generated growing opposition from local communities across the United States.

OpenAI moved pre-emptively to address these concerns at Effingham County, committing to a series of specific pledges: paying full electricity costs and infrastructure expenses rather than seeking subsidies, reducing power draw during peak grid demand periods, implementing closed-loop cooling systems designed to minimize water consumption, contributing hundreds of millions of dollars in state and local taxes, and commissioning independent third-party audits of all environmental commitments.

The last point is particularly unusual — it signals that OpenAI is aware that the credibility of corporate pledges in the AI infrastructure wave has been eroding, and is attempting to differentiate its approach with verifiable accountability mechanisms.

Context: The Enterprise Platform Play

Project Camellia is not the only major OpenAI announcement this week. The company also officially launched “Presence,” its enterprise AI agent platform, which aims to deeply integrate OpenAI models with enterprise internal data, policies, and business processes — providing automation across customer service, sales, IT support, and other functions. The platform moves OpenAI’s commercial strategy beyond selling raw model API access toward becoming the intelligence layer embedded in enterprise workflows.

Taken together, the two moves — a $30 billion infrastructure campus and an enterprise-grade agent platform — sketch OpenAI’s ambitions for the second half of this decade: own the compute, own the enterprise workflow, and collect revenue at both layers of the stack.

The Broader Infrastructure Arms Race

OpenAI’s Georgia announcement comes as every major hyperscaler is racing to secure power and land for AI infrastructure. Alphabet disclosed a potential $205 billion capex year this week. Amazon issued $25 billion in bonds to fund GPU cluster expansion. Microsoft has committed tens of billions across new facilities. Saudi Arabia’s MGX fund is deploying $49 billion into AI infrastructure globally.

The aggregate effect is a multi-trillion-dollar bet on physical infrastructure concentrated in a span of a few years — a construction boom with few historical precedents outside of wartime mobilization.

Whether the revenue to justify that spend materializes on schedule remains the central uncertainty of the current AI era. But Project Camellia makes clear that OpenAI, at least, is willing to bet its balance sheet on it.